U.S. patent application number 15/459778 was filed with the patent office on 2017-09-21 for automatic recognition of anatomical landmarks.
The applicant listed for this patent is Matthias Hoffmann, Norbert Strobel. Invention is credited to Matthias Hoffmann, Norbert Strobel.
Application Number | 20170270663 15/459778 |
Document ID | / |
Family ID | 58358399 |
Filed Date | 2017-09-21 |
United States Patent
Application |
20170270663 |
Kind Code |
A1 |
Hoffmann; Matthias ; et
al. |
September 21, 2017 |
AUTOMATIC RECOGNITION OF ANATOMICAL LANDMARKS
Abstract
A method for automatic recognition of at least one anatomical
landmark in a hollow organ of a patient is provided. The method
includes providing an image dataset of the hollow organ,
establishing or providing a three-dimensional mesh of a surface of
the hollow organ from the image dataset, and determining a
centerline of the mesh by skeletization. At least one feature is
determined for each of a plurality of points on the centerline. A
classifier pre-trained on the at least one feature is used for
detecting candidates for the at least one anatomical landmark from
the plurality of points. The candidates are grouped together with a
distance from one another below a threshold. At least one
specification determined from the anatomy of the hollow organ is
used for confirming or rejecting the candidates for the at least
one anatomical landmark. One or more candidates are defined as an
anatomical landmark.
Inventors: |
Hoffmann; Matthias;
(Nurnberg, DE) ; Strobel; Norbert; (Forchheim,
DE) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Hoffmann; Matthias
Strobel; Norbert |
Nurnberg
Forchheim |
|
DE
DE |
|
|
Family ID: |
58358399 |
Appl. No.: |
15/459778 |
Filed: |
March 15, 2017 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06T 2210/41 20130101;
G06T 17/20 20130101; G06T 2207/30101 20130101; G06T 7/0012
20130101; G06T 2207/30172 20130101; A61B 5/02007 20130101; A61B
5/7271 20130101; A61B 5/742 20130101; G06T 7/73 20170101 |
International
Class: |
G06T 7/00 20060101
G06T007/00; A61B 5/02 20060101 A61B005/02; G06T 17/20 20060101
G06T017/20; A61B 5/00 20060101 A61B005/00 |
Foreign Application Data
Date |
Code |
Application Number |
Mar 15, 2016 |
DE |
102016204225.4 |
Claims
1. A method for automatic recognition of at least one anatomical
landmark in a hollow organ of a patient, the method comprising:
providing a medical image dataset of the hollow organ; establishing
or providing a three-dimensional (3D) mesh of a surface of the
hollow organ from the medical image dataset; determining a
centerline of the mesh by skeletization; determining at least one
feature from a plurality of points on the centerline; detecting
candidates for the at least one anatomical landmark from the
plurality of points, the detecting comprising using a classifier
pre-trained on the at least one feature; grouping together the
candidates; confirming rejection of the candidates for the at least
one anatomical landmark, the confirming comprising using at least
one specification determined from the anatomy of the hollow organ;
and defining one or more candidates as an anatomical landmark.
2. The method of claim 1, wherein grouping together the candidates
comprises grouping together the candidates with a distance from one
another below a pre-specified threshold.
3. The method of claim 1, wherein the at least one feature is
formed by a feature from the group consisting of: minimum, maximum
or median of a cross-sectional surface of the centerline at the
point, in an environment, or at the point and in the environment,
spatially filtered minimum, spatially filtered maximum or spatially
filtered median of the cross-sectional surface of the centerline at
the point, in the environment or at the point and in the
environment, change of a diameter of the centerline, maximum
diameter in a distal direction, minimum diameter in a proximal
direction, position of the point in relation to the center of
gravity of the hollow organ, direction of the centerline at the
point, distance to the center of gravity of the hollow organ along
the centerline, and curvature of the surface of the mesh along the
cross-sectional surface of the centerline.
4. The method of claim 1, further comprising selecting and
outputting a suggestion dependent on the at least one anatomical
landmark for treatment planning using a classifier.
5. The method of claim 1, wherein the hollow organ is formed by a
left atrium, and the at least one anatomical landmark is formed by
a pulmonary vein ostium.
6. The method of claim 4, wherein a number of pulmonary vein ostia
are recognized.
7. The method of claim 1, wherein a further classifier is used for
recognition for each anatomical landmark to be recognized.
8. The method of claim 1, wherein one classifier is used for
recognition for a number of anatomical landmarks.
9. The method of claim 1, wherein the classifier is formed by a
decision tree.
10. The method of claim 1, wherein the 3D mesh has the form of a
triangle mesh.
11. The method of claim 4, wherein the at least one specification
includes both sides of the left atrium possessing a common ostium,
each side of the atrium possessing two pulmonary vein ostia, the
right side of the left atrium possessing two pulmonary vein ostia,
each pulmonary vein possessing only one ostium, or any combination
thereof.
12. The method of claim 1, wherein the centerline is established
such that a surface skeleton of the hollow organ is computed and
subsequently, a curve skeleton is formed from the surface
skeleton.
13. The method of claim 1, wherein providing the medical image
dataset of the hollow organ comprises forming the medical image
dataset from computed tomography image data or from magnetic
resonance tomography image data.
14. The method of claim 1, wherein the classifier is configured as
machine-learning.
15. The method of claim 1, further comprising displaying the at
least one anatomical landmark on a display unit after the defining
of the one or more candidates as an anatomical landmark.
16. The method of claim 1, further comprising computing mesh
positions that are linked to anatomical landmarks of the
centerline.
17. An apparatus for automatic recognition of at least one
anatomical landmark in a hollow organ of a patient, the apparatus
comprising: a processor configured to: provide a medical image
dataset of the hollow organ; establish or provide a
three-dimensional (3D) mesh of a surface of the hollow organ from
the medical image dataset; determine the centerline of the mesh,
the determination of the centerline of the mesh comprising
skeletization; determine at least one feature from a plurality of
points on the centerline; detect candidates for the at least one
anatomical landmark from the plurality of points, the detection of
the candidates for the at least one anatomical landmark comprising
application of a classifier pre-trained on the at least one
feature; group together the candidates with a distance from one
another below a pre-specified threshold; confirm rejection of the
candidates for the at least one anatomical landmark, the
confirmation of the rejection of the candidates comprising use of
at least one specification determined from the anatomy of the
hollow organ; and determine one or more candidates as an anatomical
landmark through the classifier; a memory configured to store the
medical image dataset; a communication unit configured to
communicate with a database; an input device configured for input
of user data; and a display configured to display a representation
of the medical image dataset.
Description
[0001] This application claims the benefit of DE 10 2016 204 225.4,
filed on Mar. 15, 2016, which is hereby incorporated by reference
in its entirety.
BACKGROUND
[0002] The present embodiments relate to automatic recognition of
at least one anatomical landmark in a hollow organ of a
patient.
[0003] In general, a marking of anatomical landmarks in medical
images is an important method of operation in order to prepare
therapy planning, for example. One example from electrophysiology
(EP) is the position of the pulmonary vein (PV) ostia and of the
left atrial appendage (LAA) in a left atrium. The marking of the
pulmonary vein ostia makes it easier to navigate catheters, since a
position of the pulmonary vein ostia may be used for the automatic
planning of catheter ablations (Koch et al., Automatic planning of
atrial fibrillation ablation lines using landmark-constrained
nonrigid registration, J. Med. Imag. 1(1), 2014), and since the
form of the pulmonary vein ostia is an important criterion in
relation to the choice of the cryo-balloon catheter to be used. A
marking of the regions and a display in an x-ray overlay image is
important for avoiding injuries caused by incorrect introduction of
the catheter.
[0004] Other examples from EP are the detection of common ostia and
the detection of the additional pulmonary veins (fifth PV). Other
applications of the proposed method are found, for example, in the
field of structural heart disease. In this field, the determination
of diameters is important (e.g., for valve replacement, for the
aorta valve), since the choice of size and possibly also type of
valve may be made accordingly. The method may likewise be used for
the detection of drainages. A further field of application is the
field of abdominal aorta aneurysms (e.g., the choice of the
appropriate stents). Additional applications are to be found in the
detection of stenoses in blood vessels and airways as well as in
the detection of fluctuations of the diameter (e.g., expansions and
constrictions) in the gastro-intestinal tract.
[0005] For therapy planning, it is usual to mark important
anatomical landmarks because anatomical landmarks are critical and
may not be constricted under any circumstances (e.g., liver
arteries in AAA therapies, coronary arteries in aorta valve
positioning, LAA for EP ablation with atrial fibrillation) or
because the anatomical landmarks are important for the positioning
of devices (EP pulmonary vein isolation). In EP in general, the
model of the left atrium is segmented using an automatic
segmentation tool from a 3D CT or MRI volume. The 3D model (e.g.,
shown as a mesh) is then displayed to the observer in an
interactive 3D view. The user marks a series of points on the
surface of the mesh to create a marking of the PV ostium and of the
LAA. Thereafter, the marked model is visualized as part of an x-ray
overlay image, through which the x-ray overlay image may then be
used to simplify the navigation within the 3D chamber.
[0006] As an alternative, a statistical model of the outline may be
used for the segmentation, with the model containing the body of
the LA and the pulmonary veins modeled individually in each case.
On account of the structure of the model, the marking of the
pulmonary vein ostia may be derived implicitly from the transition
from LA body to the pulmonary veins (e.g., from the article by
Karim et al., Left atrium pulmonary veins: segmentation and
quantification for planning atrial fibrillation ablation, Proc.
SPIE, 2009, p. 72611T ff. or from the article by Zheng et al.,
Precise segmentation of the left atrium in C-arm CT volumes with
applications to atrial fibrillation, Proc. IEEE Int. Symp. Biomed.
Imaging, IEEE, p. 1421 ff., 2012). As an alternative, a
semi-automatic approach may also be used (Rettmann et al.,
Identification of the left pulmonary vein ostia using centerline
tracking, Proc. SPIE., Vol. 7262, p. 726228ff., 2009). This method
provides that the user clicks manually on each of the four
pulmonary veins in order to calculate a centerline (e.g.,
three-dimensional (3D) central line). For each point of the
centerline, the region of the intersection of the pulmonary veins
will be calculated, and the first point, as from which the region
of the intersection becomes significantly larger, will be
considered as the ostium.
SUMMARY AND DESCRIPTION
[0007] The scope of the present invention is defined solely by the
appended claims and is not affected to any degree by the statements
within this summary
[0008] The present embodiments may obviate one or more of the
drawbacks or limitations in the related art. For example,
recognition of an anatomical landmark in a hollow organ of a
patient without any input from a user is provided.
[0009] The method for automatic recognition of at least one
anatomical landmark in a hollow organ of a patient includes
providing a medical image dataset of the hollow organ. A
three-dimensional (3D) mesh of the surface of the hollow organ is
established or provided from the image dataset. A centerline of the
mesh is determined by skeletization. At least one feature is
determined for each of a plurality of points on the centerline. A
classifier pre-trained, for example, to the at least one feature is
used for detecting candidates for the at least one anatomical
landmark from the plurality of points. The candidates with a
distance from one another below a pre-specified threshold are
grouped together. At least one specification determined from the
anatomy of the hollow organ is used for confirming or rejecting the
candidates for the at least one anatomical landmark. One or more
candidates are defined as an anatomical landmark. The advantages of
the method of one or more of the present embodiments include the
fact that intervention by a user is not needed, but all acts will
be carried out automatically. Accordingly, a user-independent,
consistent, very high quality, and improved accuracy in the
recognition of anatomical landmarks may be achieved. This provides
that the method is also very quick, so that a plurality of
detections may be carried out in a very short time, and the
clinical workflow may be speeded up. This is to the benefit of the
patient. The method is also independent of the method of recording
the 3D images, since the method does not work with volumes but with
3D meshes. For example, the method is suitable for recognizing at
least one pulmonary vein ostium (e.g., all four or five or six
pulmonary vein ostia) in a left atrium.
[0010] A centerline may be a 3D central line that has an elliptical
diameter. The publication Telea et al., Computing curve skeletons
from medial surfaces of 3D shapes, Theory and Practice of Computer
Graphics, The Eurographics Association, 2012, discloses a method
for computing such centerlines, for example. The lines are referred
to herein as curve skeletons.
[0011] According to an embodiment, the feature may be formed by a
minimum, maximum, or median of the cross-sectional surface of the
centerline at the point and/or in its environment, spatially
filtered minimum, spatially filtered maximum or spatially filtered
median of the cross-sectional surface of the centerline at the
point and/or in its environment, change of the diameter of the
centerline, maximum diameter in the distal direction, minimum
diameter in the proximal direction, position of the point in
relation to the center of gravity of the hollow organ, direction of
the centerline at the point, distance to the center of gravity of
the hollow organ along the centerline, or curvature of the surface
of the mesh along the cross-sectional surface of the centerline. A
number of features or all of the features of the centerline may be
determined. The features of the centerline may provide conclusions
about the mesh and thus about the hollow organ in a simple manner.
The determination of the features is described, inter alia, in the
published document Hoffmann et al., Automatic detection of ostia in
the left atrium, Bildverarbeitung fur die Medizin (image processing
for medicine) (BVM), Informatik Aktuell, Springer Verlag, 2016. All
or specific points in the vicinity of branches of the centerline
may be selected automatically, for example, as points for which the
features will be determined. The feature or features may also be
computed for all points on the centerline (e.g., the centerline may
be completely scanned).
[0012] According to a further embodiment, a suggestion for
treatment planning dependent on the anatomical landmark will be
selected and output by a classifier. Such a suggestion for
treatment planning may, for example, involve a suggestion for an
apparatus to be introduced in the region of the anatomical
landmark. Specifically, for example, in the case of a left atrium
and a pulmonary vein ostium, a suggestion may be output for a
catheter and/or a cryo-balloon or stent specifically adapted to the
corresponding pulmonary vein ostium. For example, this makes the
further procedure easier for a doctor, since the doctor now already
has a suggestion supplied that the doctor may then assess in
accordance with experience. The information about the catheters and
cryo-balloons (e.g., size, brand) may be retrieved from a
database.
[0013] A further classifier is used for recognition for each
anatomical landmark to be recognized. Thus, when a number of
landmarks are to be recognized, a specific classifier pre-trained
in each case for the corresponding landmark will be used. This
enables an especially high accuracy to be achieved in the
recognition of the landmarks, since each classifier is
specialized.
[0014] One classifier will be used for recognition for a number of
anatomical landmarks. The method is greatly simplified by this and
may be carried out especially quickly. The same classifier may also
be used for anatomical landmarks and associated suggestion for a
treatment plan.
[0015] According to a further embodiment, the classifier will be
formed by a decision tree. Such classifiers are known, are readily
available, are fast, and may be trained particularly well for the
corresponding landmarks. For example, a support vector machine
(SVM) or an artificial neuronal network may also be used for a
classifier. The classifier or the classifiers is/are, for example,
embodied as machine-learning. For training of the classifier or
classifiers, for example, deep-learning methods may be used. Such
methods are known, for example, from the article by Krizhevsky et
al., ImageNet classification with deep convolutional neural
networks, Advances in neural information processing systems, p.
1097 ff., 2012.
[0016] For carrying out the method in a simple manner, the
three-dimensional mesh takes the form of a triangular mesh. This
involves a widely-used and well-known form of the mesh, which is
used in computer graphics and is frequently already present for
three-dimensional medical datasets of hollow organs or may be
easily created. Other forms (e.g., rectangular meshes) may also be
used.
[0017] In accordance with a further embodiment, in the case of a
left atrium, at least one specification, including both sides of
the left atrium possessing a common ostium, each side of the atrium
possessing two pulmonary vein ostia, the right side of the left
atrium possessing two pulmonary vein ostia, each pulmonary vein
possessing only one ostium, or any combination thereof, is used to
confirm or reject candidates for anatomical landmarks from the
pre-selection. The corresponding specifications will be applied
accordingly to the candidates or to the already grouped
candidates.
[0018] According to a further embodiment, the centerline is
established such that a surface skeleton of the hollow organ is
computed, and subsequently, a curve skeleton is formed form the
surface skeleton. This is a known, simple method for determining
centerlines (e.g., described in the publication Telea et al. (see
above)).
[0019] The medical image dataset may be formed by computed
tomography image data or from magnetic resonance tomography image
data, for example.
[0020] For supporting the user or a doctor planning a therapeutic
intervention, the determination is followed by a display of the
anatomical landmarks on a display unit. The mesh positions that are
linked to the anatomical landmarks of the centerline may also be
computed.
[0021] One or more of the present embodiments also provide an
apparatus for carrying out the method. The apparatus includes a
processing unit (e.g., a processor) configured to provide a medical
image dataset of the hollow organ, establish or provide a
three-dimensional mesh of the surface of the hollow organ from the
image dataset, and determine a centerline of the mesh by
skeletization. The processor is also configured to determine at
least one feature from a plurality of points on the centerline,
apply a classifier pre-trained on the at least one feature for
detecting candidates for the at least one anatomical landmark from
the plurality of points, and collect the candidates with a distance
from one another below a pre-specified threshold. The processor is
further configured to use at least one specification determined
from the anatomy of the hollow organ for confirming or rejecting
the candidates for the at least one anatomical landmark, and
determine one or more candidates as anatomical landmark by the
classifier. The apparatus also includes a memory unit for storage
of data, a communication unit for communication with a database, an
input unit for input of user data, and a display unit for display
of image data.
BRIEF DESCRIPTION OF THE DRAWINGS
[0022] FIG. 1 shows a flowchart of one embodiment of a method;
and
[0023] FIG. 2 shows one embodiment of an apparatus configured to
carry out the method.
DETAILED DESCRIPTION
[0024] FIG. 1 shows a flowchart of one embodiment of a method with
eight acts S1 to S8. The method referring to a left atrium is shown
as an example for a hollow organ, where the aim is to recognize
pulmonary vein ostia (e.g., five: two left and three right). In act
S1, a medical three-dimensional (3D) image dataset of a left atrium
of a patient is provided (e.g., retrieved from a database or a
storage medium or retrieved directly after the corresponding
acquisition and made available). Such a three-dimensional medical
image dataset may, for example, involve a computed tomography image
dataset or a magnetic resonance tomography image dataset. In act
S2, a 3D mesh (e.g., a triangle mesh) of the surface of the volume
is computed from the image dataset. The mesh may also have been
created beforehand, and now, merely is to be retrieved from a
database or a storage medium and be made available, for example.
Such a mesh generally has nodes and edges.
[0025] In act S3 a centerline (e.g., a 3D central line), which
generally has an elliptical diameter, will be determined from the
three-dimensional mesh. Such a determination of a centerline will
generally be carried out by a skeletization. A suitable method is
described, for example, in the article by Telea et al. (see further
above), where a two-stage approach has been selected: a surface
skeleton of the left atrium (LA) is computed, and subsequently, a
curve skeleton is formed from the surface skeleton. The surface
skeleton involves a two-dimensional (2D) manifold constructed from
all the points that are the center of spheres plotted through the
mesh. The characteristic of the mesh is re-established, and a
gradient field is implicitly defined on the surface skeleton. The
points on the surface skeleton are moved iteratively along the
gradient field until the points converge along the singularities of
the gradient field, which form the curve skeleton. The result is a
shrunken mesh.
[0026] Subsequently, in a fourth act S4, at least one feature for
each of a plurality of points on the centerline is determined. In
one embodiment, the feature is formed by a minimum, maximum or
median of the cross-sectional surface of the centerline at the
point and/or its environment, spatially filtered minimum, spatially
filtered maximum or spatially filtered median of the
cross-sectional surface of the centerline at the point and/or its
environment, change of the diameter of the centerline, maximum
diameter in the distal direction, minimum diameter in the proximal
direction, position of the point in relation to the center of
gravity of the hollow organ, direction of the centerline at the
point, distance to the center of gravity of the hollow organ along
the centerline, or curvature of the surface of the mesh along the
cross-sectional surface of the centerline. A number of points or
all the points of the centerline may be determined. Points in the
vicinity of branches of the centerline may be selected
automatically, for example, as points for which the features will
be determined. The feature or features may also be computed for all
points on the centerline (e.g., the centerline may be completely
scanned).
[0027] The radius or diameter of the pulmonary veins may be
determined, for example. Generally, this is elliptical. The
cross-sectional surface may be determined at the point and the
normal vector. Subsequently, the intersection of the pulmonary
veins with this cross-sectional surface may be determined. Then,
the median of the distance from the point to the points of the
pulmonary veins produced therefrom may be determined. Thereafter, a
spatial median filtering on the radii along the pulmonary veins may
be applied. Since the pulmonary vein ostium is characterized by the
radius/diameter increasing in the direction of the center of the
left atrium, the radii or diameters of the environment of the
points will be used. To estimate the increase, the derivations of
the radius/diameter may also be used. In order to provide that a
strong increase at a point is not to be attributed to a local
minimum in the diameter of the pulmonary veins, the maximum
radius/diameter in the distal direction may also be determined. In
order to provide that the pulmonary vein ostium will not be
confused with a local expansion, the minimum radius/diameter in the
proximal direction may also be determined. In addition, the normal
vector and the distance to the center of the left atrium may be
used.
[0028] In act S5, at least one classifier pre-trained on the at
least one feature may be used for detecting candidates for the at
least one anatomical landmark from the plurality of points. The
classifier may, for example, involve a decision tree. It is
sensible to pre-train the classifier or the classifiers based on a
largest possible number of examples, so that the classifier
delivers as accurate a result as possible. The training may be
carried out, for example, by deep-learning methods. There may be
provision, for recognition of a number of pulmonary vein ostia, for
the respective ostium of each individual PV (e.g., left upper, left
lower, right upper, right lower, common and additional PV) for
using an individual classifier, but just one single classifier may
also be used for a number of pulmonary vein ostia.
[0029] In addition, a corresponding suggestion may be established
for a therapy plan from the same classifier or an individual
classifier in this context for each candidate for the landmark
(e.g., based on the anatomical circumstances). An example for
establishing candidates for pulmonary vein ostia in this context
may be a suggestion for the type/size and/or the type of a catheter
or of a cryo-balloon to be used. Suggestions for stents and other
medical equipment may also be included. This optional step is not
shown in FIG. 1.
[0030] In act S6, candidates established by the classifier or the
classifiers may be grouped together (e.g., clustered). This may be
carried out, for example, such that candidates with a distance from
one another below a previously defined threshold will be clustered
together.
[0031] In act S7, at least one specification determined from the
anatomy of the LA for confirming or rejecting candidates for the
pulmonary vein ostia is used. Such specifications may state, for
example, that only one single ostium is possible for each pulmonary
vein or that a specific side of the LA has two ostia or three ostia
(e.g., right side) or that both sides have a common ostium.
[0032] In act S8, in relation to the remaining candidates or
clusters of candidates, one or more candidates or clusters are
defined as the pulmonary vein ostium/ostia from the classifier or
the classifiers, for example. For example, the largest cluster of
candidates on the right side of the LA is defined as the first
right pulmonary vein ostium, and the largest cluster of candidates
on the left side is defined as first left pulmonary vein ostium.
The second largest cluster is then the second left pulmonary vein
ostium and the second right pulmonary vein ostium, respectively. If
there is still an ostium on the right side of the LA, the ostium of
the appendage (LAA) is involved.
[0033] Subsequently (not shown in FIG. 1), the result of the method
(e.g., the pulmonary vein ostia established) will be displayed to a
user or a doctor on a display unit. In addition (also not shown in
FIG. 1), the corresponding suggestions for therapy plans or
suggestion for therapy tools or devices to be used (e.g., type of
catheter and cryo-balloon) may be displayed.
[0034] A visualization of the anatomical landmarks with live images
from medical imaging apparatuses may subsequently be carried out.
Thus, for example, live fluoroscopy images through the triangle
mesh and the anatomical landmarks (e.g., left atrium with pulmonary
vein ostia) may be displayed overlaid.
[0035] An apparatus for carrying out the method is shown in FIG. 2.
The apparatus includes a processing unit 11 (e.g., a processor)
connected to a communication unit 9, a memory unit 10, a display
unit 12, and an input unit 13.
[0036] The present embodiments describe an automatic method for
detecting anatomical marker points that may also be used for
supporting therapy planning (e.g., by suggestions for selection of
corresponding equipment and its positioning. The atrium may be
represented as a triangle mesh, providing that the method is
independent of the modality with which the medical image dataset
has been recorded and also independent of the segmentation
tool.
[0037] One or more of the present embodiments may be briefly
summarized in the following way. For a fast and effortless support
of a user, a method for automatic recognition of at least one
anatomical landmark in a hollow organ of a patient is provided. The
method includes: providing a medical image dataset of the hollow
organ; establishing or providing a three-dimensional mesh of the
hollow organ from the image dataset; determining a centerline of
the mesh by skeletization; determining at least one feature of a
plurality of points on the centerline; using a classifier
pre-trained, for example, to the at least one feature for detecting
candidates for the at least one anatomical landmark from the
plurality of points; grouping together the candidates with a
distance from one another below a pre-specified threshold; using at
least one specification determined from the anatomy of the hollow
organ for confirming or rejecting the candidates for the at least
one anatomical landmark; and defining one or more candidates as an
anatomical landmark (e.g., by the classifier).
[0038] The elements and features recited in the appended claims may
be combined in different ways to produce new claims that likewise
fall within the scope of the present invention. Thus, whereas the
dependent claims appended below depend from only a single
independent or dependent claim, it is to be understood that these
dependent claims may, alternatively, be made to depend in the
alternative from any preceding or following claim, whether
independent or dependent. Such new combinations are to be
understood as forming a part of the present specification.
[0039] While the present invention has been described above by
reference to various embodiments, it should be understood that many
changes and modifications can be made to the described embodiments.
It is therefore intended that the foregoing description be regarded
as illustrative rather than limiting, and that it be understood
that all equivalents and/or combinations of embodiments are
intended to be included in this description.
* * * * *